All of lore.kernel.org
 help / color / mirror / Atom feed
From: "Christian König" <ckoenig.leichtzumerken@gmail.com>
To: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>,
	linaro-mm-sig@lists.linaro.org, dri-devel@lists.freedesktop.org,
	linux-media@vger.kernel.org, intel-gfx@lists.freedesktop.org
Cc: daniel@ffwll.ch
Subject: Re: [PATCH 13/26] drm/i915: use the new iterator in i915_gem_busy_ioctl
Date: Wed, 22 Sep 2021 14:22:20 +0200	[thread overview]
Message-ID: <7e03d687-4381-5263-4f03-a34f16ce8f1c@gmail.com> (raw)
In-Reply-To: <6f8140bd-9cff-e0d8-e014-b3fb8d2d8ab0@linux.intel.com>

Am 22.09.21 um 14:20 schrieb Tvrtko Ursulin:
>
> On 22/09/2021 13:15, Christian König wrote:
>> Am 22.09.21 um 13:46 schrieb Tvrtko Ursulin:
>>>
>>> On 22/09/2021 11:21, Tvrtko Ursulin wrote:
>>>>
>>>> On 22/09/2021 10:10, Christian König wrote:
>>>>> This makes the function much simpler since the complex
>>>>> retry logic is now handled else where.
>>>>>
>>>>> Signed-off-by: Christian König <christian.koenig@amd.com>
>>>>> ---
>>>>>   drivers/gpu/drm/i915/gem/i915_gem_busy.c | 35 
>>>>> ++++++++++--------------
>>>>>   1 file changed, 14 insertions(+), 21 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_busy.c 
>>>>> b/drivers/gpu/drm/i915/gem/i915_gem_busy.c
>>>>> index 6234e17259c1..313afb4a11c7 100644
>>>>> --- a/drivers/gpu/drm/i915/gem/i915_gem_busy.c
>>>>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_busy.c
>>>>> @@ -82,8 +82,8 @@ i915_gem_busy_ioctl(struct drm_device *dev, void 
>>>>> *data,
>>>>>   {
>>>>>       struct drm_i915_gem_busy *args = data;
>>>>>       struct drm_i915_gem_object *obj;
>>>>> -    struct dma_resv_list *list;
>>>>> -    unsigned int seq;
>>>>> +    struct dma_resv_iter cursor;
>>>>> +    struct dma_fence *fence;
>>>>>       int err;
>>>>>       err = -ENOENT;
>>>>> @@ -109,27 +109,20 @@ i915_gem_busy_ioctl(struct drm_device *dev, 
>>>>> void *data,
>>>>>        * to report the overall busyness. This is what the 
>>>>> wait-ioctl does.
>>>>>        *
>>>>>        */
>>>>> -retry:
>>>>> -    seq = raw_read_seqcount(&obj->base.resv->seq);
>>>>> -
>>>>> -    /* Translate the exclusive fence to the READ *and* WRITE 
>>>>> engine */
>>>>> -    args->busy = 
>>>>> busy_check_writer(dma_resv_excl_fence(obj->base.resv));
>>>>> -
>>>>> -    /* Translate shared fences to READ set of engines */
>>>>> -    list = dma_resv_shared_list(obj->base.resv);
>>>>> -    if (list) {
>>>>> -        unsigned int shared_count = list->shared_count, i;
>>>>> -
>>>>> -        for (i = 0; i < shared_count; ++i) {
>>>>> -            struct dma_fence *fence =
>>>>> -                rcu_dereference(list->shared[i]);
>>>>> -
>>>>> +    args->busy = false;
>>>>
>>>> You can drop this line, especially since it is not a boolean. With 
>>>> that:
>>>>
>>>> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
>>>
>>> Having said this, one thing to add in the commit message is some 
>>> commentary that although simpler in code, the new implementation has 
>>> a lot more atomic instructions due all the extra fence get/put.
>>>
>>> Saying this because I remembered busy ioctl is quite an over-popular 
>>> one. Thinking about traces from some real userspaces I looked at in 
>>> the past.
>>>
>>> So I think ack from maintainers will be required here. Because I 
>>> just don't know if any performance impact will be visible or not. So 
>>> view my r-b as "code looks fine" but I am on the fence if it should 
>>> actually be merged. Probably leaning towards no actually - given how 
>>> the code is localised here and I dislike burdening old platforms 
>>> with more CPU time it could be cheaply left as is.
>>
>> Well previously we would have allocated memory, which as far as I 
>> know has more overhead than a few extra atomic operations.
>
> It doesn't, it only uses dma_resv_excl_fence and dma_resv_shared_list.

Yeah, ok then that's not really an option any more.

I think Daniel and I are totally on the same page that we won't allow 
this RCU dance in the drivers any more.

Regards,
Christian.

>
> Regards,
>
> Tvrtko
>
>> On the other hand I could as well stick with dma_resv_get_fences() here.
>>
>> Regards,
>> Christian.
>>
>>>
>>> Regards,
>>>
>>> Tvrtko
>>>
>>>
>>>>
>>>> Regards,
>>>>
>>>> Tvrtko
>>>>
>>>>> + dma_resv_iter_begin(&cursor, obj->base.resv, true);
>>>>> +    dma_resv_for_each_fence_unlocked(&cursor, fence) {
>>>>> +        if (dma_resv_iter_is_restarted(&cursor))
>>>>> +            args->busy = 0;
>>>>> +
>>>>> +        if (dma_resv_iter_is_exclusive(&cursor))
>>>>> +            /* Translate the exclusive fence to the READ *and* 
>>>>> WRITE engine */
>>>>> +            args->busy |= busy_check_writer(fence);
>>>>> +        else
>>>>> +            /* Translate shared fences to READ set of engines */
>>>>>               args->busy |= busy_check_reader(fence);
>>>>> -        }
>>>>>       }
>>>>> -
>>>>> -    if (args->busy && read_seqcount_retry(&obj->base.resv->seq, 
>>>>> seq))
>>>>> -        goto retry;
>>>>> +    dma_resv_iter_end(&cursor);
>>>>>       err = 0;
>>>>>   out:
>>>>>
>>


WARNING: multiple messages have this Message-ID
From: "Christian König" <ckoenig.leichtzumerken@gmail.com>
To: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>,
	linaro-mm-sig@lists.linaro.org, dri-devel@lists.freedesktop.org,
	linux-media@vger.kernel.org, intel-gfx@lists.freedesktop.org
Cc: daniel@ffwll.ch
Subject: Re: [Intel-gfx] [PATCH 13/26] drm/i915: use the new iterator in i915_gem_busy_ioctl
Date: Wed, 22 Sep 2021 14:22:20 +0200	[thread overview]
Message-ID: <7e03d687-4381-5263-4f03-a34f16ce8f1c@gmail.com> (raw)
In-Reply-To: <6f8140bd-9cff-e0d8-e014-b3fb8d2d8ab0@linux.intel.com>

Am 22.09.21 um 14:20 schrieb Tvrtko Ursulin:
>
> On 22/09/2021 13:15, Christian König wrote:
>> Am 22.09.21 um 13:46 schrieb Tvrtko Ursulin:
>>>
>>> On 22/09/2021 11:21, Tvrtko Ursulin wrote:
>>>>
>>>> On 22/09/2021 10:10, Christian König wrote:
>>>>> This makes the function much simpler since the complex
>>>>> retry logic is now handled else where.
>>>>>
>>>>> Signed-off-by: Christian König <christian.koenig@amd.com>
>>>>> ---
>>>>>   drivers/gpu/drm/i915/gem/i915_gem_busy.c | 35 
>>>>> ++++++++++--------------
>>>>>   1 file changed, 14 insertions(+), 21 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_busy.c 
>>>>> b/drivers/gpu/drm/i915/gem/i915_gem_busy.c
>>>>> index 6234e17259c1..313afb4a11c7 100644
>>>>> --- a/drivers/gpu/drm/i915/gem/i915_gem_busy.c
>>>>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_busy.c
>>>>> @@ -82,8 +82,8 @@ i915_gem_busy_ioctl(struct drm_device *dev, void 
>>>>> *data,
>>>>>   {
>>>>>       struct drm_i915_gem_busy *args = data;
>>>>>       struct drm_i915_gem_object *obj;
>>>>> -    struct dma_resv_list *list;
>>>>> -    unsigned int seq;
>>>>> +    struct dma_resv_iter cursor;
>>>>> +    struct dma_fence *fence;
>>>>>       int err;
>>>>>       err = -ENOENT;
>>>>> @@ -109,27 +109,20 @@ i915_gem_busy_ioctl(struct drm_device *dev, 
>>>>> void *data,
>>>>>        * to report the overall busyness. This is what the 
>>>>> wait-ioctl does.
>>>>>        *
>>>>>        */
>>>>> -retry:
>>>>> -    seq = raw_read_seqcount(&obj->base.resv->seq);
>>>>> -
>>>>> -    /* Translate the exclusive fence to the READ *and* WRITE 
>>>>> engine */
>>>>> -    args->busy = 
>>>>> busy_check_writer(dma_resv_excl_fence(obj->base.resv));
>>>>> -
>>>>> -    /* Translate shared fences to READ set of engines */
>>>>> -    list = dma_resv_shared_list(obj->base.resv);
>>>>> -    if (list) {
>>>>> -        unsigned int shared_count = list->shared_count, i;
>>>>> -
>>>>> -        for (i = 0; i < shared_count; ++i) {
>>>>> -            struct dma_fence *fence =
>>>>> -                rcu_dereference(list->shared[i]);
>>>>> -
>>>>> +    args->busy = false;
>>>>
>>>> You can drop this line, especially since it is not a boolean. With 
>>>> that:
>>>>
>>>> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
>>>
>>> Having said this, one thing to add in the commit message is some 
>>> commentary that although simpler in code, the new implementation has 
>>> a lot more atomic instructions due all the extra fence get/put.
>>>
>>> Saying this because I remembered busy ioctl is quite an over-popular 
>>> one. Thinking about traces from some real userspaces I looked at in 
>>> the past.
>>>
>>> So I think ack from maintainers will be required here. Because I 
>>> just don't know if any performance impact will be visible or not. So 
>>> view my r-b as "code looks fine" but I am on the fence if it should 
>>> actually be merged. Probably leaning towards no actually - given how 
>>> the code is localised here and I dislike burdening old platforms 
>>> with more CPU time it could be cheaply left as is.
>>
>> Well previously we would have allocated memory, which as far as I 
>> know has more overhead than a few extra atomic operations.
>
> It doesn't, it only uses dma_resv_excl_fence and dma_resv_shared_list.

Yeah, ok then that's not really an option any more.

I think Daniel and I are totally on the same page that we won't allow 
this RCU dance in the drivers any more.

Regards,
Christian.

>
> Regards,
>
> Tvrtko
>
>> On the other hand I could as well stick with dma_resv_get_fences() here.
>>
>> Regards,
>> Christian.
>>
>>>
>>> Regards,
>>>
>>> Tvrtko
>>>
>>>
>>>>
>>>> Regards,
>>>>
>>>> Tvrtko
>>>>
>>>>> + dma_resv_iter_begin(&cursor, obj->base.resv, true);
>>>>> +    dma_resv_for_each_fence_unlocked(&cursor, fence) {
>>>>> +        if (dma_resv_iter_is_restarted(&cursor))
>>>>> +            args->busy = 0;
>>>>> +
>>>>> +        if (dma_resv_iter_is_exclusive(&cursor))
>>>>> +            /* Translate the exclusive fence to the READ *and* 
>>>>> WRITE engine */
>>>>> +            args->busy |= busy_check_writer(fence);
>>>>> +        else
>>>>> +            /* Translate shared fences to READ set of engines */
>>>>>               args->busy |= busy_check_reader(fence);
>>>>> -        }
>>>>>       }
>>>>> -
>>>>> -    if (args->busy && read_seqcount_retry(&obj->base.resv->seq, 
>>>>> seq))
>>>>> -        goto retry;
>>>>> +    dma_resv_iter_end(&cursor);
>>>>>       err = 0;
>>>>>   out:
>>>>>
>>


  reply	other threads:[~2021-09-22 12:22 UTC|newest]

Thread overview: 87+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-09-22  9:10 Deploying new iterator interface for dma-buf Christian König
2021-09-22  9:10 ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 01/26] dma-buf: add dma_resv_for_each_fence_unlocked v4 Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22 14:36   ` Tvrtko Ursulin
2021-09-22 14:36     ` [Intel-gfx] " Tvrtko Ursulin
2021-09-22 14:50     ` Christian König
2021-09-22 14:50       ` [Intel-gfx] " Christian König
2021-09-22 15:09       ` Tvrtko Ursulin
2021-09-22 15:09         ` [Intel-gfx] " Tvrtko Ursulin
2021-09-22  9:10 ` [PATCH 02/26] dma-buf: add dma_resv_for_each_fence Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 03/26] dma-buf: use new iterator in dma_resv_copy_fences Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 04/26] dma-buf: use new iterator in dma_resv_get_fences v2 Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 05/26] dma-buf: use new iterator in dma_resv_wait_timeout Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 06/26] dma-buf: use new iterator in dma_resv_test_signaled Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 07/26] drm/ttm: use the new iterator in ttm_bo_flush_all_fences Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 08/26] drm/amdgpu: use the new iterator in amdgpu_sync_resv Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 09/26] drm/amdgpu: use new iterator in amdgpu_ttm_bo_eviction_valuable Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 10/26] drm/msm: use new iterator in msm_gem_describe Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 11/26] drm/radeon: use new iterator in radeon_sync_resv Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 12/26] drm/scheduler: use new iterator in drm_sched_job_add_implicit_dependencies v2 Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 13/26] drm/i915: use the new iterator in i915_gem_busy_ioctl Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22 10:21   ` Tvrtko Ursulin
2021-09-22 10:21     ` [Intel-gfx] " Tvrtko Ursulin
2021-09-22 11:46     ` Tvrtko Ursulin
2021-09-22 11:46       ` [Intel-gfx] " Tvrtko Ursulin
2021-09-22 12:15       ` Christian König
2021-09-22 12:15         ` [Intel-gfx] " Christian König
2021-09-22 12:20         ` Tvrtko Ursulin
2021-09-22 12:20           ` [Intel-gfx] " Tvrtko Ursulin
2021-09-22 12:22           ` Christian König [this message]
2021-09-22 12:22             ` Christian König
2021-09-22 14:31     ` Christian König
2021-09-22 14:31       ` [Intel-gfx] " Christian König
2021-09-22 14:39       ` Tvrtko Ursulin
2021-09-22 14:39         ` [Intel-gfx] " Tvrtko Ursulin
2021-09-22  9:10 ` [PATCH 14/26] drm/i915: use the new iterator in i915_sw_fence_await_reservation v3 Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 15/26] drm/i915: use the new iterator in i915_request_await_object v2 Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22 10:34   ` Tvrtko Ursulin
2021-09-22 10:34     ` Tvrtko Ursulin
2021-09-22  9:10 ` [PATCH 16/26] drm/i915: use new iterator in i915_gem_object_wait_reservation Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 17/26] drm/i915: use new iterator in i915_gem_object_wait_priority Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 18/26] drm/i915: use new iterator in i915_gem_object_last_write_engine Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22 10:27   ` Tvrtko Ursulin
2021-09-22 10:27     ` [Intel-gfx] " Tvrtko Ursulin
2021-09-22 11:00     ` Christian König
2021-09-22 11:00       ` [Intel-gfx] " Christian König
2021-09-22 11:12       ` Tvrtko Ursulin
2021-09-22 11:12         ` [Intel-gfx] " Tvrtko Ursulin
2021-09-22  9:10 ` [PATCH 19/26] drm/i915: use new cursor in intel_prepare_plane_fb Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 20/26] drm: use new iterator in drm_gem_fence_array_add_implicit v3 Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 21/26] drm: use new iterator in drm_gem_plane_helper_prepare_fb Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 22/26] drm/nouveau: use the new iterator in nouveau_fence_sync Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 23/26] drm/nouveau: use the new interator in nv50_wndw_prepare_fb Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 24/26] drm/etnaviv: use new iterator in etnaviv_gem_describe Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 25/26] drm/etnaviv: replace dma_resv_get_excl_unlocked Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22  9:10 ` [PATCH 26/26] dma-buf: nuke dma_resv_get_excl_unlocked Christian König
2021-09-22  9:10   ` [Intel-gfx] " Christian König
2021-09-22 12:55 ` [Intel-gfx] ✗ Fi.CI.BUILD: failure for series starting with [01/26] dma-buf: add dma_resv_for_each_fence_unlocked v4 Patchwork
  -- strict thread matches above, loose matches on Subject: below --
2021-09-21 17:36 [PATCH 01/26] dma-buf: add dma_resv_for_each_fence_unlocked v3 Christian König
2021-09-21 17:36 ` [PATCH 13/26] drm/i915: use the new iterator in i915_gem_busy_ioctl Christian König
2021-09-17 12:34 Deploying new iterator interface for dma-buf Christian König
2021-09-17 12:35 ` [PATCH 13/26] drm/i915: use the new iterator in i915_gem_busy_ioctl Christian König
2021-09-16 11:30 Deploying new iterator interface for dma-buf Christian König
2021-09-16 11:30 ` [PATCH 13/26] drm/i915: use the new iterator in i915_gem_busy_ioctl Christian König
2021-09-13 13:16 Deploying new iterator interface for dma-buf Christian König
2021-09-13 13:16 ` [PATCH 13/26] drm/i915: use the new iterator in i915_gem_busy_ioctl Christian König

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=7e03d687-4381-5263-4f03-a34f16ce8f1c@gmail.com \
    --to=ckoenig.leichtzumerken@gmail.com \
    --cc=daniel@ffwll.ch \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=intel-gfx@lists.freedesktop.org \
    --cc=linaro-mm-sig@lists.linaro.org \
    --cc=linux-media@vger.kernel.org \
    --cc=tvrtko.ursulin@linux.intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.